execute_child shouldn't use any signal-unsafe functions

[yasushi-saito]

I see the uses of std::string and setenv after fork. I believe these functions are unsafe, because after fork, the state of memory heap is undefined. I haven't been able to produce crash myself so this issue isn't super critical, but it'd be good to fix them.

For setenv, perhaps we should create a new environ before fork in the parent. For std::string, we can just write a raw string using write(2) and call _exit.

[nextsilicon-itay-bookstein]

I've run into this as a bug. On auditing the code, I see that there are also exception throws inside the sensitive span of code between fork and exec. As exception throws may allocate, they should also be avoided. I'll see if I can arrange a patch for this.

[nextsilicon-itay-bookstein]

#120